Enhancing Breast Cancer Diagnosis Through Clustering: A Study of KMeans, Agglomerative, and Gaussian Mixture Models

Abstract

In the evolving landscape of medical data analysis, clustering techniques play a pivotal role, particularly in deciphering intricate patterns within datasets, such as those linked to cancer diagnostics. With the continuous expansion and increasing complexity of healthcare data, there is a growing demand for effective clustering algorithms capable of extracting significant insights. Current trends underscore the necessity of carefully selecting the most appropriate clustering method to improve both the accuracy and interpretability of analytical results. In this paper, we conduct a comprehensive comparison of three prominent clustering algorithms - KMeans, Agglomerative Clustering, and Gaussian Mixture Model (GMM) - applied to a breast cancer dataset comprising features from Fine Needle Aspirates (FNA) of breast masses. Following a thorough preprocessing and scaling of the features, we assess the performance of these clustering techniques using the Silhouette Score, Calinski-Harabasz Score, and Davies- Bouldin Score. The findings reveal that KMeans provides superior cluster separation and clarity relative to the other algorithms. This research emphasizes the critical role of algorithm selection based on specific dataset attributes and evaluation metrics, aiming to enhance the accuracy of clustering outcomes in breast cancer classification.